Early warning of low visibility using the ensembling of machine learning approaches for aviation services at Jay Prakash Narayan International (JPNI) Airport Patna

Abstract

AbstractExtremely low visibility affects aviation services. Aviation services need accurate fog and low-visibility predictions for airport operations. Fog and low-visibility forecasting are difficult even with modern numerical weather prediction models and guiding systems. Limitations in comprehending the micro-scale processes that lead to fog formation, intensification, onset, and dissipation complicate fog prediction. This article predicts low visibility for Jay Prakash Narayan International Airport (JPNI), Patna, India, using a historical synoptic dataset. The proposed machine learning (ML) approaches optimize three meta-algorithm approaches: boosting (which reduces variances), bagging (which reduces bias), and stacking (which improves predictive forces). The ML approaches optimize the best prediction algorithms (at level 0) for fog (surface visibility ≤ 1000 m) and dense fog (surface visibility ≤ 200 m), and the suggested ensemble models at level 1 (an ensemble of level 0 ML approaches) deliver the highest performance and stability in prediction output. All time series perform well with the specified model (6-h to 1-h lead time for any combination of observed historical datasets). Airport management, planning, and decision-making rely on high reliability. Because it works well and is reliable, the proposed approaches can be used at other airports in India's Indo-Gangetic Plain.